Presented By: Michelle Rusch, University of Illinois College of Medicine at Peoria
Co-Authors: Sean Creeden, University of Illinois College of Medicine at Peoria
Daniel Henley, University of Illinois College of Medicine at Peoria
Inki Kim, University of Illinois Urbana-Champaign
Jane Maksimovic, University of Illinois College of Medicine at Peoria
Claudia Mello-Thoms, University of Iowa
AJ Pool, University of Illinois College of Medicine at Peoria
Andy Tu, University of Illinois College of Medicine at Peoria

Purpose 
Residents in radiology typically suffer greater effects of fatigue compared with attendings. Differences are likely the result of novice readers using a selective feature-based exhaustive search opposed to a non-selective gestalt approach. Cumulative effects of related fatigue can lead to error. This study examines the question, does a head mounted augmented reality display help or hinder abnormality detection and offset fatigue in novice radiologists compared to expert attendings? 

Methods 
Eight radiologists (4 attendings, 4 residents) provided a preliminary review of the testing interface (on MagicLeap [Ver 2.0, Magic Leap Inc., Plantation FL] and a standard monitor) to be used in a pilot study. The prototype presented abnormality detection in six CT head cases in videos presented at six frames per second (normal, subdural hematoma, subarachnoid hemorrhage, primary brain tumor, stroke). Effects of fatigue were examined using self report (SOFI) and objective measures (Critical Flicker Fusion).  

Results 
Feedback related to interface usability on both displays was mostly favorable (93%). MagicLeap ratings were also positive for control inputs and lighting (83%). There was variability in perceptions of the importance of video rates and scrolling (reported both in scores and comments). Forty percent of respondents considered the videos to be too fast. Analysis of relationships between fatigue and accuracy are still in progress.  

Conclusion 
Feedback from the review provided lessons learned to inform the design for a pilot study (n=12) testing 32 CT head imaging sets. The pilot interface will integrate flexibility in imaging presentation (e.g., slower rates, images in reverse) to evaluate eye-tracking, detection accuracy, and speed. A region selection task will be added to increase statistical power. Future findings have the potential to positively affect accurate interpretation in radiology and provide the foundations for a novel collaborative clinical AI tool.